mgmt/glusterd: do cleanup of snapshots in post-validate phase if half baked

               objects are there

Change-Id: I372cac98ad054cdc1a6fbc7f6c77c25981063b2f
Signed-off-by: Raghavendra Bhat <raghavendra@redhat.com>
Reviewed-on: http://review.gluster.org/7237
Reviewed-by: Rajesh Joseph <rjoseph@redhat.com>
Tested-by: Rajesh Joseph <rjoseph@redhat.com>

1 files changed, 57 insertions, 0 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-store.c b/xlators/mgmt/glusterd/src/glusterd-store.c
index ae20b60db..613f03abc 100644
--- a/xlators/mgmt/glusterd/src/glusterd-store.c
+++ b/xlators/mgmt/glusterd/src/glusterd-store.c
@@ -3030,6 +3030,35 @@ out:
         return ret;
 }
 
+int32_t
+glusterd_resolve_snap_bricks (xlator_t  *this, glusterd_snap_t *snap)
+{
+        int32_t                 ret = -1;
+        glusterd_volinfo_t      *volinfo = NULL;
+        glusterd_brickinfo_t    *brickinfo = NULL;
+
+        GF_ASSERT (this);
+        GF_VALIDATE_OR_GOTO (this->name, snap, out);
+
+        list_for_each_entry (volinfo, &snap->volumes, vol_list) {
+                list_for_each_entry (brickinfo, &volinfo->bricks, brick_list) {
+                        ret = glusterd_resolve_brick (brickinfo);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "resolve brick failed in restore");
+                                goto out;
+                        }
+                }
+        }
+
+        ret = 0;
+
+out:
+        gf_log (this->name, GF_LOG_TRACE, "Returning with %d", ret);
+
+        return ret;
+}
+
 int
 glusterd_store_update_snap (glusterd_snap_t *snap)
 {
@@ -3152,6 +3181,34 @@ glusterd_store_retrieve_snap (char *snapname)
                 goto out;
         }
 
+        /* Unlike bricks of normal volumes which are resolved at the end of
+           the glusterd restore, the bricks belonging to the snap volumes of
+           each snap should be resolved as part of snapshot restore itself.
+           Because if the snapshot has to be removed, then resolving bricks
+           helps glusterd in understanding what all bricks have its own uuid
+           and killing those bricks.
+        */
+        ret = glusterd_resolve_snap_bricks (this, snap);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING, "resolving the snap bricks"
+                        " failed (snap: %s)", snap?snap->snapname:"");
+
+        /* When the snapshot command from cli is received, the on disk and
+           in memory structures for the snapshot are created (with the status)
+           being marked as GD_SNAP_STATUS_INIT. Once the backend snapshot is
+           taken, the status is changed to GD_SNAP_STATUS_IN_USE. If glusterd
+           dies after taking the backend snapshot, but before updating the
+           status, then when glusterd comes up, it should treat that snapshot
+           as a failed snapshot and clean it up.
+        */
+        if (snap->snap_status != GD_SNAP_STATUS_IN_USE) {
+                ret = glusterd_snap_remove (snap, _gf_true, _gf_true);
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING, "failed to remove"
+                                " the snapshot %s", snap->snapname);
+                goto out;
+        }
+
         /* TODO: list_add_order can do 'N-square' comparisions and
            is not efficient. Find a better solution to store the snap
            in order */